Inference Index
All models
Directory · Models · AI21
Jamba

Jamba 2

Hybrid Mamba + Transformer architecture. Linear cost scaling on long contexts, so the 256K window is actually fast.

Input / 1M
$0.50
Output / 1M
$1.50
Context
256K
Released
Nov 14, 2025

Benchmark profile

MMLU-ProHumanEvalIFEvalTruthfulQAHarmBenchLiveBench
Profile
MMLU-Pro75.5%
HumanEval77.0%
IFEval83.0%
TruthfulQA83.0%
HarmBench83.0%
LiveBench75.5%

Scores

Editorial

Strengths
  • +Linear-time long context
  • +Cheap long-doc RAG
  • +Open weights
Trade-offs
  • Smaller ecosystem of fine-tunes

Price history (last 90 days)

History starts accumulating on the next nightly snapshot.

What this costs to use

All task cards →

Recommended alternatives